diff --git a/aidge_export_cpp/kernels/transpose.hpp b/aidge_export_cpp/kernels/transpose.hpp
new file mode 100644
index 0000000000000000000000000000000000000000..31c9e27869c5e2fde701f6700fd4964ea4cefd29
--- /dev/null
+++ b/aidge_export_cpp/kernels/transpose.hpp
@@ -0,0 +1,87 @@
+/********************************************************************************
+ * Copyright (c) 2023 CEA-List
+ *
+ * This program and the accompanying materials are made available under the
+ * terms of the Eclipse Public License 2.0 which is available at
+ * http://www.eclipse.org/legal/epl-2.0.
+ *
+ * SPDX-License-Identifier: EPL-2.0
+ *
+ ********************************************************************************/
+
+#ifndef __AIDGE_EXPORT_CPP_KERNELS_TRANSPOSE__
+#define __AIDGE_EXPORT_CPP_KERNELS_TRANSPOSE__
+
+/**
+ * @brief Transposes an N-dimensional tensor based on the specified permutation.
+ *
+ * This function rearranges the dimensions of an N-dimensional tensor according to the
+ * permutation array provided. The input tensor is expected to have dimensions specified
+ * by `in_dims`, and the output tensor will have dimensions reordered as specified by the
+ * `permute` array.
+ *
+ * Based on Tensor::copyTranspose from aidge.aidge_core
+ *
+ * @tparam T Data type of the tensor elements.
+ * @tparam NB_DIMS Number of dimensions of the input tensor.
+ * @param[in] inputs Pointer to the input tensor data stored in contiguous memory.
+ * @param[in] in_dims Array containing the size of each dimension of the input tensor.
+ * @param[in] permute Array of unsigned integers specifying the desired permutation
+ * of dimensions. Each value should be in the range [0, NB_DIMS-1],
+ * defining the new order of dimensions for the output tensor.
+ * @param[in] total_size Total number of elements in the input/output tensor.
+ * @param[out] outputs Pointer to the pre-allocated memory for the transposed tensor.
+ * Ensure this memory is appropriately sized to hold the transposed data.
+ */
+template <typename T,unsigned int NB_DIMS>
+void transpose_ND_forward(const T *__restrict inputs,
+ const unsigned int *in_dims,
+ const unsigned int *permute,
+ const unsigned int total_size,
+ T *__restrict outputs)
+{
+ // Compute strides for input tensor
+ unsigned int in_strides[NB_DIMS];
+ in_strides[NB_DIMS - 1] = 1;
+ for (int i = NB_DIMS - 2; i >= 0; --i)
+ {
+ in_strides[i] = in_strides[i + 1] * in_dims[i + 1];
+ }
+
+ // Compute dimensions and strides for output tensor
+ unsigned int out_dims[NB_DIMS];
+ unsigned int out_strides[NB_DIMS];
+ out_strides[NB_DIMS - 1] = 1;
+ for (unsigned int i = 0; i < NB_DIMS; ++i)
+ {
+ out_dims[i] = in_dims[permute[i]];
+ }
+ for (int i = NB_DIMS - 2; i >= 0; --i)
+ {
+ out_strides[i] = out_strides[i + 1] * out_dims[i + 1];
+ }
+
+ unsigned int current_idx[NB_DIMS];
+
+ // Iterate over all elements in the input tensor
+ for (unsigned int idx = 0; idx < total_size; ++idx)
+ {
+
+ unsigned int remaining = idx;
+ for (unsigned int i = 0; i < NB_DIMS; ++i)
+ {
+ current_idx[i] = remaining / in_strides[i];
+ remaining = remaining % in_strides[i];
+ }
+
+ unsigned int output_index = 0;
+ for (unsigned int i = 0; i < NB_DIMS; ++i)
+ {
+ output_index += current_idx[permute[i]] * out_strides[i];
+ }
+
+ outputs[output_index] = inputs[idx];
+ }
+}
+
+#endif // __AIDGE_EXPORT_CPP_KERNELS_TRANSPOSE__
diff --git a/aidge_export_cpp/operators.py b/aidge_export_cpp/operators.py
index f04dbb3bd134fbb517f677649b05032e5944f886..54c38055b2c622aa1796bc1e7ff2dd46e60afcba 100644
--- a/aidge_export_cpp/operators.py
+++ b/aidge_export_cpp/operators.py
@@ -57,7 +57,7 @@ class ProducerCPP(ExportNode):
self.values = np.array(self.operator.get_output(0))
if len(self.values.shape) == 4: # Note: export in HWC
- self.values = np.transpose(self.values, (0, 2, 3, 1))
+ self.values = np.transpose(self.values, (0, 2, 3, 1))
def export(self, export_folder: Path):
header_path = f"include/parameters/{self.attributes['name']}.h"
@@ -295,3 +295,16 @@ class FcCPP(ExportNodeCpp):
str(ROOT / "kernels" / "activation.hpp"),
str(ROOT / "kernels" / "rescaling.hpp")
]
+
+@ExportLibCpp.register("Transpose", aidge_core.ImplSpec(aidge_core.IOSpec(aidge_core.dtype.any)))
+class TransposeCPP(ExportNodeCpp):
+ def __init__(self, node, mem_info):
+ super().__init__(node, mem_info)
+ self.config_template = str(
+ ROOT / "templates" / "configuration" / "transpose_ND_config.jinja")
+ self.forward_template = str(
+ ROOT / "templates" / "kernel_forward" / "transpose_ND_forward.jinja")
+ self.include_list = []
+ self.kernels_to_copy = [
+ str(ROOT / "kernels" / "transpose.hpp")
+ ]
\ No newline at end of file
diff --git a/aidge_export_cpp/templates/configuration/transpose_ND_config.jinja b/aidge_export_cpp/templates/configuration/transpose_ND_config.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..e5ef4ffbf8ced740f00b5ba716348bc9cc06ca8c
--- /dev/null
+++ b/aidge_export_cpp/templates/configuration/transpose_ND_config.jinja
@@ -0,0 +1,15 @@
+{#- For name header -#}
+#ifndef {{ name|upper }}_LAYER_H
+#define {{ name|upper }}_LAYER_H
+
+{# For layer configuration -#}
+{% include "./_def_io.jinja" %}
+{% include "./_meminfo.jinja" %}
+{# Export suppose that batchsize = 1#}
+#define {{ name|upper }}_NB_DIMS {{ in_dims[0] | length }}
+
+static constexpr unsigned int {{ name|upper }}_PERMUTE[] = { {{ output_dims_order | join(', ') }} };
+static constexpr unsigned int {{ name|upper }}_DIMS[] = { {{ in_dims[0] | join(', ') }}};
+
+
+#endif /* {{ name|upper }}_LAYER_H */
\ No newline at end of file
diff --git a/aidge_export_cpp/templates/kernel_forward/transpose_ND_forward.jinja b/aidge_export_cpp/templates/kernel_forward/transpose_ND_forward.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..25af5bd9a3cdab4c91d5f2f09dae9144348729db
--- /dev/null
+++ b/aidge_export_cpp/templates/kernel_forward/transpose_ND_forward.jinja
@@ -0,0 +1 @@
+transpose_ND_forward<{{in_cdtype[0]}},{{name|upper}}_NB_DIMS>({{in_name[0]}},{{name|upper}}_DIMS,{{name|upper}}_PERMUTE,{{ out_name[0]|upper }}_SIZE,{{out_name[0]}});
\ No newline at end of file